基于图迁移学习的梯级水电发电超短期预测 |
点此下载全文 |
引用本文:张海锋1,杨晶莹1,冷俊1,魏泽涛2,沈晓东2.基于图迁移学习的梯级水电发电超短期预测[J].电网与清洁能源,2023,39(10):104~112 |
摘要点击次数: 134 |
全文下载次数: 67 |
|
基金项目:国家自然科学基金联合基金项目(U22B20123);国网吉林省电力有限公司科技项目(2022JBGS-04) |
|
中文摘要:水电是可再生能源的重要组成部分,精确预测水电站的发电量对电力系统的运行和调度至关重要。针对传统预测方法在处理水电站之间复杂的拓扑结构时存在限制的问题,提出了一种基于图迁移学习的方法,旨在通过水电站的拓扑连接特征提升发电预测的准确性和泛化能力。利用水电站的拓扑结构构建图表示各水电站之间的关联关系,以捕捉水电站的上下游关联特征,采用预训练源水电站数据集并通过图迁移学习来适应目标水电站的发电预测模型。实验结果表明,图迁移学习有助于更好地捕捉水电站间的拓扑特征,提高发电预测精度,减少所需训练样本数量。 |
中文关键词:迁移学习 图神经网络 参数迁移 水电发电预测 |
|
Ultra-Short-Term Forecasting of Cascaded Hydropower Generation Based on Graph Transfer Learning |
|
|
Abstract:Hydropower is an essential component of renewable energy sources, and accurately predicting the power generation of hydropower stations is crucial for the operation and scheduling of the power system. To address the limitations of traditional forecasting methods in dealing with the complex topological structures among hydropower stations, a graph transfer learning-based approach is proposed, aiming at enhancing the accuracy and generalization capability of power generation prediction by leveraging the topological connectivity features of hydropower stations. A graph is constructed using the topological structure of hydropower stations to represent the relationships between them, capturing the upstream and downstream associations among the stations. Pre-training is performed on a source hydropower station dataset, and then graph transfer learning is employed to adapt the power generation prediction model to the target hydropower station. The experimental results show that graph transfer learning helps to better capture the topological features among hydropower stations, leading to improved accuracy in power generation prediction and a reduced requirement for training samples. |
keywords:transfer learning graph neural networks parameter migration hydropower generation prediction |
查看全文 查看/发表评论 下载PDF阅读器 |